Integrated operation parameter measurement and control system for box-type transformer

Abstract

The present application relates to the technical field of power equipment comprehensive measurement and diagnosis, and particularly relates to a comprehensive operation parameter measurement and control system for box-type transformers, which comprises a multi-domain parameter acquisition module, a time-space alignment and fusion module, a dynamic window abnormality diagnosis module, a cross-domain coupling degree calculation module and a twin self-tuning module.After the multi-domain parameter acquisition module acquires multi-source heterogeneous data, it generates a synchronous parameter sequence through time-space alignment and fusion; based on the information entropy deviation, the dynamic window abnormality diagnosis module constructs an abnormality criterion under the electrical domain overrun trigger; if the criterion is abnormal, the cross-domain coupling degree calculation module represents the electrical-thermal-insulation chain reaction strength, and the twin self-tuning module completes online parameter correction.The present application completely covers the electrical-thermal-insulation chain delay process through an adaptive time window, and effectively improves the multi-domain coupling fault identification accuracy and reduces missed judgments and false alarms through digital twin closed-loop parameter adjustment.

Description

Technical Field

[0001] This application relates to the field of integrated measurement and diagnostic technology for power equipment, and in particular to an integrated operation parameter measurement and control system for box-type transformers. Background Technology

[0002] As a core device at the end of the distribution network, prefabricated transformers generally adopt an integrated multi-compartment structure comprising a high-voltage compartment, a transformer compartment, and a low-voltage compartment. The cabinet houses instrument transformers, temperature sensors, partial discharge sensors, and access control switches, simultaneously collecting four types of operational signals: electrical, thermal, insulation, and security. Existing monitoring systems collect signals from each sensor area in an isolated manner, resulting in inherent heterogeneity in the sampling period and timing characteristics of various signals: electrical loads are continuous signals at the millisecond level, thermal environment parameters are slowly varying signals at the second / minute level, and partial discharge pulses and access control position changes are instantaneous discrete step signals without fixed intervals.

[0003] Current monitoring systems generally use a uniform linear interpolation method to complete the time-series synchronization of all data. This synchronization method does not take into account the differentiated physical characteristics of electrical continuous signals, partial discharge pulses and access control change discrete step signals. During the interpolation smoothing process, the instantaneous change characteristics of partial discharge pulses and the boundary of access control status jumps are easily weakened, resulting in a deviation between the synchronized data and the actual dynamic changes of multiple physical fields inside the transformer, which affects the validity of the basic data for subsequent equipment status analysis.

[0004] Transformer faults exhibit a chain-like evolution pattern, with electrical overload, heat accumulation, and insulation deterioration propagating step by step. Changes in thermal and insulation states caused by electrical parameters exceeding limits have a significant lag effect, and existing monitoring and analysis methods are insufficient to fully track this type of cross-physical field hysteresis evolution process. Summary of the Invention

[0005] To address the problems of distortion in the unified linear interpolation synchronization data of existing monitoring systems and the difficulty in fully tracking the hysteretic chain evolution process of electrical-thermal-insulation after electrical limits are exceeded, this invention provides solutions in the following aspects.

[0006] A comprehensive operation parameter measurement and control system for box-type transformers includes: a multi-domain parameter acquisition module for acquiring multi-source heterogeneous data of the box-type transformer during operation, wherein the multi-source heterogeneous data includes: electrical continuous signals, discrete switching events and thermal environment step signals; The spatiotemporal alignment and fusion module is used to perform classification spatiotemporal alignment and hot cumulative feature extraction on multi-source heterogeneous data, and output aligned and synchronized data with a unified time benchmark. The dynamic window anomaly diagnosis module is configured to take electrical domain limit overrun as the trigger condition, dynamically adjust the analysis time window width according to the amplitude of electrical stress mutation, and construct anomaly criteria based on the deviation between the multi-domain information entropy within the window and the historical steady-state information entropy. The cross-domain coupling degree calculation module is configured to calculate the cross-domain coupling degree when a potential anomaly is determined. The cross-domain coupling degree is used to characterize the intensity of the chain reaction of anomalies in the electrical-thermal-insulation multi-domain domains. It is determined based on the maximum cross-correlation coefficient of the historical sequence using the time lag difference and is verified and updated online. The twin self-tuning module is configured to calibrate initial parameters through an offline digital twin model and adaptively correct the time window width and anomaly criterion threshold online based on the feedback of the cross-domain coupling degree and information entropy deviation. After acquiring heterogeneous data from multiple sources, the multi-domain parameter acquisition module generates a synchronous parameter sequence through spatiotemporal alignment and fusion. Under the triggering of electrical domain limit exceedance, the dynamic window anomaly diagnosis module constructs anomaly criteria based on information entropy deviation. If an anomaly is determined, the intensity of the electrical-thermal-insulation chain reaction is characterized by cross-domain coupling degree calculation, and the parameters are corrected online by the twin self-tuning module.

[0007] Optionally, the calculation logic for dynamically adjusting the analysis time window width based on the amplitude of electrical stress mutation is as follows: The electrical domain parameters are monitored in real time. When a voltage or current surge amplitude is detected that exceeds the preset stress baseline, an anomaly diagnosis process is triggered. Based on the deviation of the surge amplitude from the stress baseline, combined with the preset reference time window width and adaptive scaling factor, the dynamically adjusted analysis time window width is obtained by linear scaling calculation.

[0008] Optionally, the steps for constructing the anomaly criterion are as follows: Parameters from the electrical domain, thermal environment domain, and insulation state domain are extracted from the aligned and synchronized data and dimensionless normalization is performed to construct a multi-domain joint feature vector. After symbolic processing of the feature vector, the joint information entropy within the window is calculated. The absolute deviation between the current information entropy and the offline calibrated historical steady-state information entropy is calculated. When the deviation is greater than the preset anomaly criterion threshold, it is determined that there is a potential anomaly in the current time period.

[0009] Optionally, the method for calculating the cross-domain coupling degree is as follows: By combining the ratio of the time-series change of thermal accumulation characteristics to the amplitude of electrical stress mutation and the exponential decay term corresponding to the time lag difference, the cross-domain coupling degree is obtained. The maximum cross-correlation coefficient between the electrical stress sequence and the thermal environment response sequence is calculated offline based on historical dynamic operating data, and the duration corresponding to the maximum cross-correlation coefficient is taken as the initial time lag difference. During online operation, the peak times of the temperature rise curve and the electrical load integral curve are continuously matched to complete the verification and update of the time lag difference.

[0010] Optionally, the twin self-tuning module is specifically configured as follows: An offline digital twin model of thermal-circuit coupling for a box-type transformer is established. Historical fault data under all operating conditions is used to initially calibrate the reference time window width, adaptive scaling factor, and anomaly criterion threshold. During the online operation phase, the cross-domain coupling degree and information entropy deviation output in real time are used as inputs to generate correction coefficients, and the reference time window width, adaptive scaling factor, and anomaly criterion threshold are updated synchronously.

[0011] Optionally, the process of calibrating the initial parameters using an offline digital twin model is as follows: The parameter ranges of the baseline time window width, adaptive scaling factor, and anomaly criterion threshold are traversed, and the anomaly diagnosis and cross-domain coupling degree calculation process is performed in batch simulations. The initial calibration values ​​of each parameter are determined based on the screening criteria of the highest anomaly detection rate, the lowest false alarm rate, and the stability of numerical calculation under small disturbances, and the safety margin verification is completed.

[0012] Optionally, it also includes an abnormal link location and alarm push module, which, after determining the abnormality and completing the cross-domain coupling degree calculation, identifies the abnormal propagation path where the coupling strength between electrical-thermal-insulation exceeds the threshold; generates standardized abnormal link tags containing the time of the abnormality, the fault source domain, the coupling propagation link, and the coupling strength level; pushes the tags and alarm data to the monitoring and maintenance terminal via industrial Ethernet or wireless communication, and visualizes the multi-domain abnormal propagation links.

[0013] Optionally, the alignment and synchronization data of the unified time base includes: For continuous electrical signals, timestamp linear interpolation is used for resampling; for discrete switching events and thermal environment step signals, zero-order hold method is used for resampling; taking the physical geometric center of the transformer core and windings as the heat source center, the weighting coefficient is determined according to the reciprocal of the straight Euclidean distance from each temperature sensor to the heat source center, the collected temperature values ​​are weighted and summed to extract thermal accumulation features, and the processed data is mapped to the same time grid to generate a synchronization parameter sequence.

[0014] This application has the following technical effects: 1. This invention achieves time synchronization by differentiating the physical characteristics of continuous electrical signals, discrete partial discharge pulses, and access control step signals. Continuous signals are interpolated linearly, while discrete / step signals are prohibited from interpolation and are aggregated using time windows and zero-order hold resampling. Simultaneously, spatial distance reciprocal weighted fusion is combined to extract aggregated thermal accumulation features to compensate for environmental temperature differences. This avoids the pulse waveform loss and step state blurring and distortion caused by traditional uniform interpolation, and fully preserves the original physical characteristics of various electrical, thermal, insulation, and security signals. It generates time-uniform, physically distortion-free aligned synchronization data, providing a reliable data source for cross-domain chain evolution analysis.

[0015] 2. This invention uses a linear scaling analysis window based on the amplitude of electrical stress mutations. The window automatically widens as the overload exceeds the limit, which can fully cover the multi-physics hysteresis evolution cycle under all operating conditions from slight disturbances to severe overloads. At the same time, through offline cross-correlation calibration and online iterative updating of time lag difference through peak matching, combined with cross-domain coupling degree calculation with minimum value protection term and time lag decay correction, it can avoid calculation overflow under small stresses and suppress pseudo-coupling caused by long-term random superposition. It can also quantitatively characterize the strength of the chain reaction of temperature rise and insulation degradation induced by electrical disturbances, and realize complete tracking and quantitative evaluation of the entire hysteresis evolution process.

[0016] 3. This invention uses a digital twin model to traverse parameter ranges to complete the optimal offline calibration of window width, adaptive coefficient, and anomaly threshold, while simultaneously verifying numerical stability and safety margin. During online operation, it dynamically corrects the entire set of diagnostic parameters based on real-time information entropy deviation and cross-domain coupling feedback, compensating for the model adaptability degradation caused by seasonal temperature changes, cooling aging, and sensor drift. Combined with multi-domain joint symbolic information entropy criteria and anomaly link location and push module, it significantly improves the ability to detect early latent anomalies, reduces false alarms caused by normal disturbances, and ensures the accuracy of electrical-thermal-insulation chain anomaly identification and tracing in the long term. Attached Figure Description

[0017] Figure 1 This is a module connection diagram of the integrated operation parameter measurement and control system for box-type transformers according to an embodiment of the present invention. Detailed Implementation

[0018] The invention discloses a comprehensive operation parameter measurement and control system for box-type transformers, with reference to... Figure 1 It includes: a multi-domain parameter acquisition module, a spatiotemporal alignment and fusion module, a dynamic window anomaly diagnosis module, a cross-domain coupling degree calculation module, and a twin self-tuning module. The multi-domain parameter acquisition module acquires multi-source heterogeneous data and generates a synchronization parameter sequence through spatiotemporal alignment and fusion. The dynamic window anomaly diagnosis module constructs anomaly criteria based on information entropy deviation under electrical domain limit exceedance triggering. If an anomaly is determined, the cross-domain coupling degree is calculated to characterize the intensity of the electrical-thermal-insulation chain reaction, and the twin self-tuning module completes online parameter correction.

[0019] The multi-domain parameter acquisition module is used to acquire multi-source heterogeneous data of the box-type transformer during operation. The multi-source heterogeneous data includes: continuous electrical signals, discrete switching events and thermal environment step signals.

[0020] The spatiotemporal alignment and fusion module is used to perform classification spatiotemporal alignment and hot cumulative feature extraction on multi-source heterogeneous data, and output aligned and synchronized data with a unified time benchmark.

[0021] The box-type transformer is equipped with a high-voltage chamber, a transformer chamber, and a low-voltage chamber. Various types of sensors are distributed in each chamber, and the sampling frequency and output signal characteristics of different sensor nodes are significantly different.

[0022] Specifically, voltage transformers and current transformers collect continuous electrical load parameters using millisecond-level sampling; temperature sensors and humidity sensors collect thermal parameters such as oil temperature and ambient temperature and humidity using second-level or minute-level sampling; all the above-mentioned collected data are equipped with timestamps obtained through time synchronization via the IEEE 1588 PTP protocol or NTP protocol.

[0023] Ultra-high frequency (UHF) and high-frequency partial discharge sensors output partial discharge pulse signals, belonging to the insulation state sensing domain; access control microswitches output access control position change signals, belonging to the physical security sensing domain. Both types of signals are discrete events or step signals, and the events are instantaneous and unpredictable. The data format is a sparse pulse sequence with non-equal interval time scales or a level transition record.

[0024] If a uniform interpolation algorithm is used to perform timing alignment for all signals without distinction, it is easy to cause the loss of the original waveform information of discrete pulses and the blurring and distortion of the step signal shape, which in turn leads to physical feature distortion and timing information misalignment in the subsequent state analysis process.

[0025] In response to the heterogeneous characteristics of multi-domain parameters in the temporal and spatial domains, the processing engine performs classification and spatiotemporal alignment processing on the acquired original multi-domain parameters.

[0026] For continuous signals such as electrical load and temperature that change approximately continuously over time, there is generally no abrupt change in physical quantities between adjacent sampling points. Time alignment is achieved by using linear interpolation based on timestamps.

[0027] In practice, a unified reference synchronization time sequence is selected. For each reference time, the two sets of time-stamped sampled data closest to each continuous signal before and after that reference time are retrieved. Linear interpolation is performed by setting weights according to the timestamp difference ratio to obtain the corresponding continuous signal synchronization sampled value at that reference time. Linear interpolation is a well-known technique in this field and will not be elaborated here. The continuous signal sequence aligned in the above manner can maintain the original physical change trend and will not introduce step distortion.

[0028] For discrete event signals represented by partial discharge pulses and step signals represented by access control position changes, these signals have the inherent characteristics of instantaneous triggering and state transition at the physical level. Therefore, time window aggregation or zero-order hold resampling is used for processing, and linear interpolation is prohibited for discrete event signals and step signals.

[0029] For discrete event signals such as partial discharge pulses, a time window aggregation method is adopted: a sliding time window with the same length as the reference synchronization period is set, the number of partial discharge pulses falling into each time window or the cumulative amount of pulse amplitude is counted, and the statistical results are used as the alignment feature value of the corresponding reference time. Timing synchronization is completed while retaining the event statistical information.

[0030] For step signals such as access control position changes, a zero-order hold resampling method is adopted: at each reference time, the access control state corresponding to the nearest previous actual sampling time is retrieved as the synchronization value of the current reference time; the synchronization state value is not updated until the next actual position change event occurs. This processing method can ensure that the step edge of the resampling result is consistent with the actual position change time, avoiding the problem of virtual and real distortion of the access control state caused by smoothing processing.

[0031] Within the thermal environment sensing domain, temperature sensors are arranged in different spatial locations within the transformer room, and the linear Euclidean distances from each temperature sensor to the physical geometric center of the transformer core and windings are different for each other.

[0032] Due to the influence of heat conduction path and heat dissipation conditions, the closer the measuring point is to the center of the heat source, the stronger the temperature value is indicative of the true thermal state of the iron core and winding; the farther the measuring point is from the center of the heat source, the more easily the temperature value is affected by environmental convection and local low temperature regions.

[0033] To extract the thermal accumulation features that characterize the overall thermal state of the transformer while filtering out local noise at the measurement points, a feature extraction method based on inverse spatial distance attenuation weighting is adopted.

[0034] First, the physical geometric center of the transformer core and windings is defined as the heat source center. The straight-line Euclidean distance from the three-dimensional installation position of each temperature sensor to the heat source center is measured in meters.

[0035] The closer the temperature sensor measuring point is to the center of the heat source, the higher the weight of the measuring point temperature in representing the thermal state of the transformer body. The spatial weight coefficient is calculated by taking the reciprocal of the distance from a single sensor measuring point to the center of the heat source, dividing it by the sum of the reciprocals of the distances of all temperature sensors participating in the calculation inside the transformer room, thus achieving weight normalization. The sum of the weight coefficients corresponding to all temperature sensors equals 1.

[0036] After determining the weighting coefficients for each temperature sensor, the cumulative evolution characteristics of the aggregate heat are calculated: the temperature sampling values ​​of each temperature sensor at the current moment after time-series alignment are multiplied by the corresponding weighting coefficients and accumulated, and then the environmental background temperature compensation factor is superimposed.

[0037] Specifically, the evolution characteristics of the cumulative polymerization heat satisfy the following relationship: ; In the formula, Indicates in The evolution characteristics of the cumulative heat of aggregation at each time point; This indicates the total number of temperature sensors involved in the calculation of thermal accumulation characteristics. Indicates the first Spatial thermal gradient weighting coefficients corresponding to each temperature sensor; Indicates the first A temperature sensor in Temperature sample values ​​after time-series alignment, in Kelvin (kJ / ... ); This represents the ambient background temperature compensation factor, which is determined based on the long-term monitoring average of the transformer room ventilation opening or the external ambient temperature. An empirical value of 2.5K is used, but it can also be adjusted by the implementer according to the specific implementation scenario.

[0038] In the long-term temperature and humidity measurement statistics and no-load steady-state temperature rise calibration test of outdoor box-type substations, the ventilation structure of the box, the heat insulation conditions and the installation position of the measuring points will cause the temperature measurement data to form a stable static temperature difference of 2.5K relative to the external reference environment. This value is selected as the middle value of the industry-standard 2-3K compensation range, which is suitable for various working conditions. Setting this compensation factor can eliminate the fixed measurement offset caused by the ambient base temperature, so that the cumulative heat accumulation characteristics only reflect the actual heat generation temperature rise of the transformer body, unify the thermal characteristic evaluation benchmark under different installation scenarios, avoid the calculation deviation introduced by environmental interference, and ensure the diagnostic accuracy of subsequent electro-thermal cross-domain coupling analysis.

[0039] In this embodiment, the multi-domain parameters can be divided into the electrical stress domain, the thermal environment domain, the insulation state domain, and the physical security domain.

[0040] The aggregated heat accumulation evolution characteristic represents the overall thermal state characterization of the transformer at the current moment, with the unit being Kelvin; relying on the above spatial weighted fusion method, it can effectively suppress the local temperature disturbance introduced by sensor measuring points far from the heat source center.

[0041] After the aforementioned classification and spatiotemporal alignment processing, the original continuous signals such as electrical load and temperature are converted into continuous sampling sequences with a unified time scale; partial discharge pulses are converted into time-synchronized pulse statistical characteristic sequences; access control status signals are resampled into time-synchronized state step sequences; and temperature data from each measuring point are fused to generate a cumulative thermal evolution characteristic sequence. All types of signals are uniformly mapped to the same reference time scale, resulting in complete aligned and synchronized data.

[0042] Aligned and synchronized data can preserve the original physical change trend of continuous signals and retain the inherent abrupt change characteristics of discrete events and step signals. At the same time, by using the spatial thermal gradient weighted fusion method, local noise in the thermal environment domain is suppressed, providing a standardized data foundation with spatiotemporal consistency for subsequent anomaly diagnosis and cross-domain coupling analysis.

[0043] The dynamic window anomaly diagnosis module is configured to use electrical domain limit exceedance as the trigger condition, dynamically adjust the analysis time window width according to the amplitude of electrical stress mutation, and construct anomaly criteria based on the deviation between the multi-domain information entropy within the window and the historical steady-state information entropy.

[0044] When the information entropy deviation is greater than the anomaly criterion threshold, it is determined that there is a potential anomaly within the current time window, and the aligned synchronization data within the window is sent into the cross-domain causal coupling degree calculation process for subsequent alarm source tracing analysis; otherwise, it is determined that it is only a normal operation disturbance and there is no need to carry out subsequent cross-domain coupling analysis.

[0045] After acquiring alignment synchronization data, the processing leads continuously monitor electrical stress exceeding the threshold.

[0046] Within the electrical stress domain, the effective value of the current on either the high-voltage or low-voltage side of the transformer is selected as the monitoring object, and an electrical over-limit reference value is set. This initial reference value is jointly calibrated by an offline digital twin model in conjunction with the transformer's rated current and equipment operating procedures. An exemplary value is 1.05 times the rated current, in amperes, and online adaptive adjustment is supported during operation.

[0047] When the real-time calculated effective current value is higher than the electrical over-limit reference value, the event-driven analysis process is triggered, and the current current value is recorded as the electrical stress mutation amplitude, in amperes.

[0048] The severity of electrical over-limit events varies, resulting in significant differences in the hysteresis duration of the induced thermal accumulation response and the insulation condition response. Using a fixed-length analysis window cannot simultaneously address the need for effective capture of cross-domain correlation information under both minor disturbance and severe overload conditions. To adapt to the hysteresis response period under different overload intensities, this solution dynamically adjusts the analysis time window width based on the amplitude of the electrical stress mutation.

[0049] The baseline time window width and adaptive scaling factor are preset, and the initial values ​​of the two parameters are calibrated by the offline digital twin model. For example, the baseline time window width is 5.0 seconds and the adaptive scaling factor is 1.0. Both parameters support online self-tuning optimization.

[0050] The dynamic time window width is calculated using the following logic: based on the baseline time window width, the window expansion amount caused by the relative degree of exceeding the limit is superimposed; the relative degree of exceeding the limit is determined by the ratio of the absolute amplitude of the stress mutation amplitude exceeding the limit baseline value to the limit baseline value, and then combined with the adaptive scaling factor to obtain the window expansion ratio.

[0051] Specifically, the width of the dynamic time window satisfies the following relationship: ; In the formula, Indicates the width of the dynamic time window, in seconds (s); Indicates the width of the reference time window; Indicates the adaptive scaling factor; Indicates the amplitude of sudden changes in electrical stress; This indicates the electrical over-limit reference value.

[0052] The adaptive proportional coefficient is used to linearly map the relative electrical overload level to a time window expansion rate, controlling the rate at which the dynamic time window widens with the overload amplitude. The initial value of this coefficient is determined through simulation calibration of multiple overload conditions using a digital twin model of the box-type transformer, combined with the measured lag time of field faults. The value needs to consider two constraints: firstly, ensuring that the window can be fully extended under severe overload scenarios to completely capture the electrical-thermal-insulation chain lag response process; secondly, avoiding excessive window expansion under slight disturbances to prevent invalid data from increasing the system's computational burden. In this implementation, an exemplary value of 1.0 is used. This initial parameter is suitable for the overload range of conventional distribution network box-type transformers and supports online self-tuning based on historical anomaly samples during system operation. It can be flexibly optimized according to equipment heat dissipation conditions, terminal computing power, and field conditions.

[0053] The greater the stress exceeds the baseline limit, the wider the dynamic time window becomes, thus ensuring that the analysis window can fully cover the entire hysteresis evolution process corresponding to the electrical-thermal-insulation chain reaction on the time scale.

[0054] After determining the width of the dynamic time window, the electrical stress sequence, the cumulative evolution characteristic sequence of polymer heat, and the statistical characteristic sequence of partial discharge pulses within the time window range are extracted from the alignment and synchronization data.

[0055] Different physical quantities have significantly different dimensions. To eliminate the analytical interference caused by the dimensions, normalization is performed on each sequence to construct normalized feature vectors for the electrical, thermal, and insulation domains at each time point. ,in, This represents the normalized features of the electrical dimension. This represents the normalized characteristics of the thermal environment dimension. This represents the normalized feature of the insulation dimension.

[0056] The normalization implementation is as follows: The electrical stress value is divided by the transformer's rated stress reference to obtain the electrical dimension normalized characteristic. The rated stress reference can be selected as the transformer's nameplate rated current or rated apparent power. The cumulative evolution characteristic of aggregated heat is divided by the steady-state thermal equilibrium temperature rise reference value to obtain the thermal dimension normalized characteristic. The steady-state thermal equilibrium temperature rise reference value is the long-term steady-state average of the cumulative evolution characteristic of aggregated heat under rated load conditions. The partial discharge pulse statistical characteristic is divided by the upper limit of partial discharge statistics to obtain the insulation dimension normalized characteristic. The upper limit of partial discharge statistics can be selected as the maximum observed value of this parameter during the equipment's historical operation or the offline calibration value. All the above normalized characteristics are dimensionless values.

[0057] The time window contains several sampling moments, and each sampling moment generates a corresponding three-domain normalized feature vector. To solve for the information entropy of the multi-domain feature vectors within the window, this implementation method employs a joint symbolic processing method.

[0058] Each dimension of the normalized feature vector is divided into a predetermined number of intervals of equal width; for example, an empirical value of 10 is chosen. For the feature vector at any given time, the interval indices corresponding to the intervals to which each dimension of the vector belongs are obtained. These three sets of interval indices are then concatenated to obtain the joint symbol corresponding to that sampling time. , Indicates the first The joint symbol corresponding to each sampling time; This represents the interval index corresponding to the normalized feature of the electrical dimension; This represents the interval index corresponding to the normalized feature of the thermal environment dimension; This represents the interval index corresponding to the normalized feature of the insulation dimension.

[0059] The frequency of occurrence of each type of joint symbol in all samples within the current time window is counted. The probability of occurrence of each type of joint symbol is calculated based on the total number of samples. The information entropy within the window is then calculated based on the symbol probabilities.

[0060] Information entropy is used to characterize the joint uncertainty of the normalized operating states of the electrical, thermal environment, and insulation domains within the current window. The more dispersed the equipment operating state patterns within the window, the larger the information entropy value, indicating that the degree to which the multi-physics coupling relationship deviates from the stable operating mode during that period is higher.

[0061] Historical steady-state information entropy benchmarks are pre-obtained during the offline calibration phase: multiple steady-state time windows with normal equipment operation and no electrical over-limit events are selected, and the information entropy is calculated separately according to the same symbolization and information entropy calculation process. This scheme does not simply use the mean as the benchmark, but instead conducts high-frequency sampling of the information entropy of a large number of steady-state samples, and obtains the stable distribution center through Gaussian distribution fitting or kernel density estimation, and determines this distribution center as the steady-state information entropy benchmark.

[0062] During online diagnostics, the absolute deviation between the current window's information entropy and the steady-state information entropy benchmark is calculated. This deviation quantifies the degree of deviation of the current window's multi-domain operating state from its historical steady-state operating distribution. The obtained information entropy deviation is then compared with a preset anomaly criterion threshold.

[0063] The initial value of the anomaly criterion threshold is calibrated by the offline digital twin model, with an exemplary empirical value of 0.5 Knights. At the same time, the system can also adaptively adjust the threshold online based on historical data of cross-domain coupling degree and information entropy deviation.

[0064] When the information entropy deviation exceeds the anomaly criterion threshold, a potential anomaly is determined to exist within the current time window. The start and end times of the analysis window and the information entropy deviation value are recorded, and the aligned and synchronized data within the window is sent to the cross-domain causal coupling degree calculation process. Conversely, if the deviation is less than the threshold, it is determined to be merely a normal operation disturbance, and no further cross-domain coupling analysis is required.

[0065] The cross-domain coupling degree calculation module is configured to calculate the cross-domain coupling degree when a potential anomaly is determined. The cross-domain coupling degree is used to characterize the intensity of the chain reaction of anomalies in the electrical-thermal-insulation multi-domain system. It is determined based on the maximum cross-correlation coefficient of the historical sequence using the time lag difference and is verified and updated online.

[0066] When the aforementioned steps determine that the information entropy deviation within the current dynamic time window exceeds the anomaly criterion threshold, the joint uncertainty representing the normalized operating state of the electrical, thermal, and insulation domains significantly deviates from the historical steady-state distribution. An abnormal coupling evolution process between the electrical, thermal, and insulation domains is highly likely to occur within the time window. To quantitatively characterize the strength of the multiphysics chain reaction corresponding to this potential anomaly, the processing engine performs cross-domain causal coupling degree calculations within the current time window.

[0067] From a physical mechanism perspective: after the electrical domain generates a stress mutation that acts on the transformer body, the cumulative heat effect and insulation performance degradation response both exhibit objective time lags. Therefore, the electrical parameter time series and the thermal characteristic time series cannot be directly aligned for calculation. Before incorporating the aggregated thermal accumulation characteristic sequence and the electrical stress sequence into the causal strength calculation, it is necessary to pre-determine an effective time correspondence benchmark between the two time series.

[0068] To this end, multiple sets of continuous operating sequence data containing typical overload, light load, and normal start-stop conditions were retrieved from the historical dynamic operating condition database. The normalized cross-correlation coefficient between the electrical stress sequence and the cumulative thermal characteristic sequence was calculated by sliding along the time series. The cross-correlation coefficient varied with the number of sliding offset steps, and the lag time corresponding to the maximum value of the cross-correlation coefficient was selected as the initial time lag difference. This initial time lag difference reflects the average delay scale of the transmission of electrical stress disturbance to the point of generating thermal response under the combined effects of transformer structure, cooling conditions, and the field operating environment.

[0069] During the online operation of the system, in order to avoid parameter inaccuracies caused by sensor drift and changes in load characteristics, the effectiveness of the time lag difference is continuously verified.

[0070] The specific verification method is as follows: For each electrical limit violation and window data with potential anomalies determined by information entropy deviation, calculate the peak time of the electrical load integral curve and the peak time of the local gradient of the aggregate temperature rise curve; calculate the difference between the two peak times. If the deviation between this difference and the currently used time lag difference does not exceed the preset time verification tolerance, for example, a relative deviation of 20% is selected as the tolerance, then the current time lag difference remains unchanged; if it exceeds the tolerance range, then the average of the peak time differences corresponding to the most recent set number of limit violation events is used to update the time lag difference, realizing online parameter iteration.

[0071] After determining the effective time lag difference, the processing engine performs cross-domain causal coupling degree calculation within the time window for anomaly detection.

[0072] Within this time window, the amplitude of electrical stress mutation has been recorded in the previous steps. At the same time, the cumulative thermal accumulation feature sequence within the window is extracted from the alignment and synchronization data, and the maximum change of the thermal accumulation feature within the window relative to the start time of the window is calculated.

[0073] The amplitude of stress mutation exceeding the over-limit reference is defined to characterize the intensity of electrical disturbance. Under low load fluctuations or slight over-limit conditions, this amplitude tends to zero. Directly using it as the denominator in calculations can easily lead to calculation overflow. Therefore, a pre-defined minimal constant is added to the denominator as a protection term. An example value is provided. The unit should be consistent with the amplitude of the electrical disturbance.

[0074] Specifically, the degree of cross-domain causal coupling satisfies the following relationship: ; In the formula, Indicates the degree of cross-domain causal coupling; This represents the maximum change in the cumulative heat of polymerization characteristic within the window, expressed in Kelvin. This indicates the magnitude by which electrical stress exceeds the over-limit reference, expressed in amperes. Represents a minimal constant; This represents the time lag difference after online verification and update, in seconds. This represents the time constant of the transformer's thermal response characteristics, measured in seconds. It is used for normalizing the lag time and can be determined based on the transformer model and factory temperature rise test data. A typical value is 3600 seconds.

[0075] The cross-domain causal coupling degree is determined by two factors: the first is the ratio of the maximum change in thermal accumulation to the amplitude of the electrical disturbance after the addition of a minimum protection term. This factor characterizes the thermal response amplitude induced by a unit change in electrical stress, reflecting the sensitive coupling strength between the electrical domain and the thermal environment domain. When heat dissipation conditions and insulation material conditions remain stable, this ratio remains within a relatively fixed range; when the internal cooling channels of the transformer are blocked, the insulation medium ages, or a local hot spot fault occurs, the same electrical disturbance will produce a larger temperature rise, and this ratio will increase accordingly.

[0076] The second term is an exponential decay term, used to normalize the time delay between electrical disturbances and thermal response. It introduces a characteristic time constant corresponding to the transformer's thermal condition, mapping the time lag difference to a dimensionless value. If the time lag difference is much larger than the characteristic time constant, the exponential term decays significantly. Its physical meaning is that two sets of parameters with excessively large lag times are weakly correlated or even have no direct causal relationship at the physical level. This term suppresses the pseudo-coupling phenomenon caused by random superposition over long time scales.

[0077] The larger the cross-domain causal coupling degree value, the higher the intensity of the electrical-thermal-insulation chain reaction within the current anomaly window, and the more prominent the deep coupling characteristics of multi-physics fields. This calculation result will serve as the data basis for subsequent anomaly tracing and fault criterion analysis.

[0078] The twin self-tuning module is configured to calibrate initial parameters through an offline digital twin model and adaptively adjust the time window width and anomaly criterion threshold online based on feedback from cross-domain coupling degree and information entropy deviation.

[0079] The alarm and location unit is configured to generate multi-domain coupling anomaly link tags and push them outward based on anomaly criteria and cross-domain causal coupling degree calculation results. Specifically, when the aforementioned steps determine that a multi-domain coupling anomaly exists, the alarm and location unit extracts the coupling degree matrix corresponding to the anomaly time, identifies anomaly propagation paths in the electrical domain, thermal environment domain, and insulation partial discharge domain where the coupling strength exceeds a preset threshold; generates standardized link tags, the tag content of which includes the anomaly type, involved device nodes, coupling strength level, and timestamp, and pushes the alarm information to the monitoring platform or operation and maintenance terminal through the communication interface.

[0080] During the offline calibration phase, multi-domain synchronous data accumulated from the long-term operation of the box-type transformer are collected. The dataset covers various load levels, ambient temperatures, cooling mode switching conditions, and known minor abnormal conditions. A digital twin model is constructed based on the above dataset.

[0081] This digital twin model is built based on thermal-circuit coupling equivalent parameters. After inputting electrical load sequence, ambient temperature sequence and initial insulation state parameters, it can numerically simulate and output the characteristics of aggregated heat accumulation, the statistical trend of partial discharge pulse changes, and the information entropy distribution under the corresponding operating conditions.

[0082] The spatiotemporal alignment processing unit is configured to perform classified spatiotemporal alignment processing on the acquired raw monitoring data: for continuous signals, time linear interpolation is used; for discrete events and step signals, time window aggregation or zero-order hold resampling is used; in the thermal environment domain, aggregated thermal accumulation features are extracted based on the spatial distance inverse decay weighted algorithm, and finally aligned synchronization data is generated.

[0083] The event-driven analysis unit is configured to execute the aforementioned online analysis logic, complete the anomaly criterion solution and cross-domain causal coupling degree calculation; the self-tuning unit is configured to adaptively adjust the time window related parameters and anomaly criterion thresholds based on the information entropy deviation and coupling degree statistical characteristics stored in the sliding record stack; the alarm and location unit is configured to generate multi-domain coupling anomaly link tags and complete information push.

[0084] The offline calibration process is implemented as follows: First, select multiple historical time windows whose duration is an integer multiple of the baseline time window and which have been confirmed by the operation logs to be in a normal steady-state operating state; extract aligned synchronization data within each steady-state window and calculate the information entropy using the aforementioned joint symbolization method. During the offline calibration process, the background baseline information entropy is used to obtain a stable distribution center through distribution fitting, thereby obtaining the historical steady-state information entropy baseline and ensuring that the steady-state information entropy baseline is representative of the operating conditions.

[0085] The steady-state information entropy benchmark is increased by a certain factor to obtain the initial value of the anomaly criterion threshold; at the same time, the initial values ​​of the benchmark time window width and the adaptive scaling factor are temporarily given. Based on this, multiple sets of historical time series samples containing electrical over-limit events, and subsequent observations of actual temperature rise and partial discharge increments, are input into the digital twin model; using a step-by-step traversal method, various parameter combinations of the benchmark time window width, adaptive scaling factor, and anomaly criterion threshold are traversed within the preset value range.

[0086] For each set of parameters, the entire online analysis process was reproduced in the digital twin simulation environment. The detection rate of real abnormal events and the false triggering rate of normal operation disturbances were statistically analyzed based on the information entropy deviation and cross-domain causal coupling degree under the current parameters. A small number of minor over-limit disturbance samples with amplitudes only exceeding the electrical over-limit benchmark were injected to verify that the minimum value protection term in the denominator of the coupling degree calculation formula remained effective, effectively preventing numerical overflow.

[0087] By comprehensively screening parameter combinations, the combination with the highest detection rate, lowest false trigger rate, and stable and abrupt coupling degree calculation results under minor disturbance conditions is selected as the offline calibration parameters for the system. This step simultaneously performs safety margin verification to ensure that the calibration parameters do not exhibit numerical calculation anomalies or defects such as time window truncation of the electrical-thermal-insulation chain lag evolution process under extreme operating conditions.

[0088] After the system is put into online operation, the fixed parameters obtained from offline calibration will gradually lose their adaptability to operating conditions due to factors such as seasonal fluctuations in ambient temperature, slow degradation of transformer cooling efficiency, and long-term drift of sensor measurement links. Therefore, after each limit-crossing event that is determined to be a potential anomaly by online analysis, the system continuously stores the corresponding information entropy deviation and cross-domain causal coupling degree data into a fixed-length sliding record stack.

[0089] Once the number of samples in the stack reaches the set capacity, an online parameter tuning evaluation is initiated for each newly added record. First, the average entropy deviation and the average cross-domain causal coupling degree of all information in the stack are calculated.

[0090] If multiple consecutive evaluations meet the following conditions: the average information entropy deviation is below the first lower bound and the average coupling degree is below the lower limit of coupling degree, it means that the current parameter combination is too sensitive to normal slight disturbances and is prone to generating false alarms; at this time, the anomaly criterion threshold is adjusted upward by a set step size.

[0091] Conversely, if multiple consecutive evaluations meet the following conditions: the average information entropy deviation is higher than the first upper bound, and the average coupling degree is higher than the upper limit of coupling degree, it indicates that the current criterion threshold is too high, and there is a risk of missing real anomalies. In this case, the anomaly criterion threshold will be adjusted downward by a set step size. The upper and lower bound values ​​are preset based on the statistical quantiles obtained during the offline calibration phase.

[0092] The online self-tuning rules for the reference time window width and adaptive scaling factor are as follows: After each set number of evaluation cycles, the first-order difference trend of the coupling degree sequence within the stack is statistically analyzed. The difference between the average coupling degree of two adjacent evaluation cycles is calculated. If the difference is greater than the positive trend threshold, it indicates that the internal heat conduction path of the transformer has deteriorated or the cooling performance has decreased, and the duration of the thermal response caused by the same electrical stress disturbance has become longer. The original time window parameters cannot fully cover the extended hysteresis evolution process, and the reference time window width is expanded proportionally. If the difference is less than the negative trend threshold, the reference time window width is slightly reduced in the opposite direction.

[0093] The adaptive scaling factor usually remains unchanged from the offline calibration value; only when the cumulative adjustment of the reference time window exceeds 30% of the original calibration value will the adaptive scaling factor be scaled in conjunction with the dynamic time window to maintain the inherent scaling and adjustment characteristics of the dynamic time window.

[0094] This system is also equipped with an alarm and location unit. The alarm and location unit is configured to generate multi-domain coupling anomaly link tags and push them outward based on adaptive anomaly criteria and cross-domain causal coupling degree calculation results. When a multi-domain coupling anomaly is determined to exist, the alarm and location unit extracts key parameters of the multi-domain during the anomaly period, traces the source of the anomaly in reverse, generates standardized tags containing the anomaly time, anomaly source domain, propagation path, and scope of impact, and pushes them to the upper-level monitoring system via industrial Ethernet or wireless communication. The anomaly propagation link is also visualized on the human-machine interface to help maintenance personnel locate the root cause of the fault.

[0095] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A comprehensive operation parameter measurement and control system for box-type transformers, characterized in that, include: The multi-domain parameter acquisition module is used to acquire multi-source heterogeneous data of the box-type transformer during operation. The multi-source heterogeneous data includes: electrical continuous signals, discrete switching events and thermal environment step signals. The spatiotemporal alignment and fusion module is used to perform classification spatiotemporal alignment and hot cumulative feature extraction on multi-source heterogeneous data, and output aligned and synchronized data with a unified time benchmark. The dynamic window anomaly diagnosis module is configured to take electrical domain limit overrun as the trigger condition, dynamically adjust the analysis time window width according to the amplitude of electrical stress mutation, and construct anomaly criteria based on the deviation between the multi-domain information entropy within the window and the historical steady-state information entropy. The cross-domain coupling degree calculation module is configured to calculate the cross-domain coupling degree when a potential anomaly is determined. The cross-domain coupling degree is used to characterize the intensity of the chain reaction of anomalies in the electrical-thermal-insulation multi-domain domains. It is determined based on the maximum cross-correlation coefficient of the historical sequence using the time lag difference and is verified and updated online. The twin self-tuning module is configured to calibrate initial parameters through an offline digital twin model and adaptively correct the time window width and anomaly criterion threshold online based on the feedback of the cross-domain coupling degree and information entropy deviation. After acquiring heterogeneous data from multiple sources, the multi-domain parameter acquisition module generates a synchronous parameter sequence through spatiotemporal alignment and fusion. Under the triggering of electrical domain limit exceedance, the dynamic window anomaly diagnosis module constructs anomaly criteria based on information entropy deviation. If an anomaly is determined, the intensity of the electrical-thermal-insulation chain reaction is characterized by cross-domain coupling degree calculation, and the parameters are corrected online by the twin self-tuning module.

2. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The calculation logic for dynamically adjusting the analysis time window width based on the amplitude of electrical stress mutation is as follows: The electrical domain parameters are monitored in real time. When a voltage or current surge amplitude is detected that exceeds the preset stress baseline, an anomaly diagnosis process is triggered. Based on the deviation of the surge amplitude from the stress baseline, combined with the preset reference time window width and adaptive scaling factor, the dynamically adjusted analysis time window width is obtained by linear scaling calculation.

3. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The steps for constructing the anomaly criterion are as follows: Extract electrical, thermal environment, and insulation state parameters from the aligned synchronization data and perform dimensionless normalization to construct a multi-domain joint feature vector. After symbolizing the feature vectors, the joint information entropy within the window is calculated; the absolute deviation between the current information entropy and the historical steady-state information entropy calibrated offline is calculated; when the deviation is greater than the preset anomaly criterion threshold, it is determined that there is a potential anomaly in the current time period.

4. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The method for calculating cross-domain coupling degree is as follows: By combining the ratio of the time-series change of thermal accumulation characteristics to the amplitude of electrical stress mutation and the exponential decay term corresponding to the time lag difference, the cross-domain coupling degree is obtained. The maximum cross-correlation coefficient between the electrical stress sequence and the thermal environment response sequence is calculated offline based on historical dynamic operating data, and the duration corresponding to the maximum cross-correlation coefficient is taken as the initial time lag difference. During online operation, the peak times of the temperature rise curve and the electrical load integral curve are continuously matched to complete the verification and update of the time lag difference.

5. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The twin self-tuning module is specifically configured as follows: An offline digital twin model of thermal-circuit coupling for a box-type transformer was established, and the initial calibration of the baseline time window width, adaptive scaling factor, and anomaly criterion threshold was completed using historical fault data under all operating conditions. During the online operation phase, the cross-domain coupling degree and information entropy deviation output in real time are used as inputs to generate correction coefficients, and the baseline time window width, adaptive scaling factor and anomaly criterion threshold are updated synchronously.

6. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The process of calibrating initial parameters using an offline digital twin model is as follows: Traverse the parameter ranges of the baseline time window width, adaptive scaling factor, and anomaly criterion threshold, and perform batch simulation to execute the anomaly diagnosis and cross-domain coupling degree calculation process; The initial calibration values ​​of each parameter were determined based on the screening criteria of the highest anomaly detection rate, the lowest false alarm rate, and the stability of numerical calculation under small disturbances, and the safety margin verification was completed.

7. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, It also includes an abnormal link location and alarm push module, which identifies abnormal propagation paths where the coupling strength between electrical-thermal-insulation exceeds the threshold after determining the abnormality and completing the cross-domain coupling degree calculation; generates standardized abnormal link tags containing the time of abnormality occurrence, fault source domain, coupling propagation link, and coupling strength level; pushes the tags and alarm data to the monitoring and maintenance terminal through industrial Ethernet or wireless communication, and visualizes the multi-domain abnormal propagation links.

8. The integrated operation parameter measurement and control system for box-type transformers according to claim 1, characterized in that, The alignment and synchronization data of the unified time base includes: For continuous electrical signals, timestamp linear interpolation is used for resampling; for discrete switching events and thermal environment step signals, zero-order hold method is used for resampling; taking the physical geometric center of the transformer core and windings as the heat source center, the weighting coefficient is determined according to the reciprocal of the straight Euclidean distance from each temperature sensor to the heat source center, the collected temperature values ​​are weighted and summed to extract thermal accumulation features, and the processed data is mapped to the same time grid to generate a synchronization parameter sequence.

Images